Hi David,
I can at least confirm the issue on ab.
And the stress can be reduced a lot for better debugg-ability:

$ ab -k -n 1 -c 1 https://DavidFavor.com/foo-8192.txt | grep '^Time per 
request.*mean)$'
Time per request:       397.659 [ms] (mean)
$ ab -k -n 1 -c 1 https://DavidFavor.com/foo-8193.txt | grep '^Time per 
request.*mean)$'
Time per request:       5400.170 [ms] (mean)

So one byte makes the request x13 longer.

But as one single request shows the same issue, we can compare much easier:
$ time wget https://DavidFavor.com/foo-8192.txt --quiet
real    0m0.416s
$ time wget https://DavidFavor.com/foo-8193.txt --quiet
real    0m0.416s

So yeah, something in ab most likely.

I checked that with the version in Artful to not face a known issue -
but it still is occuring there.

Checking with "strace -rT" reveals that it is just about as fast with 8193 than 
8192.
Except there is a last final call on epoll_wait with 5 seconds.

That together with many definitions in ab being on 8192 I'd assume it
fails to understand that it is done by exceeding some buffer and then
opens up a new epoll which is on a 5 second timeout.

That is clearly an upstream problem (or they consider >8192 not supported).
Would you mind reporting there and posting the bug reference here then?
That would be very kind.
In terms of an "ubuntu bug" this is "confirmed" but not really actionable.

Thanks btw for creating an online setup one can reuse to test this!

** Changed in: apache2 (Ubuntu)
       Status: New => Confirmed

-- 
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/1704168

Title:
  Apache-2.4.27 massive slow down for files >8192 bytes long

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/apache2/+bug/1704168/+subscriptions

-- 
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs

Reply via email to